Subunit-selective N-Methyl-d-aspartate (NMDA) Receptor Signaling through Brefeldin A-resistant Arf Guanine Nucleotide Exchange Factors BRAG1 and BRAG2 during Synapse Maturation*

The maturation of glutamatergic synapses in the CNS is regulated by NMDA receptors (NMDARs) that gradually change from a GluN2B- to a GluN2A-dominated subunit composition during postnatal development. Here we show that NMDARs control the activity of the small GTPase ADP-ribosylation factor 6 (Arf6) by consecutively recruiting two related brefeldin A-resistant Arf guanine nucleotide exchange factors, BRAG1 and BRAG2, in a GluN2 subunit-dependent manner. In young cortical cultures, GluN2B and BRAG1 tonically activated Arf6. In mature cultures, Arf6 was activated through GluN2A and BRAG2 upon NMDA treatment, whereas the tonic Arf6 activation was not detectable any longer. This shift in Arf6 regulation and the associated drop in Arf6 activity were reversed by a knockdown of BRAG2. Given their sequential recruitment during development, we examined whether BRAG1 and BRAG2 influence synaptic currents in hippocampal CA1 pyramidal neurons using patch clamp recordings in acute slices from mice at different ages. The number of AMPA receptor (AMPAR) miniature events was reduced by depletion of BRAG1 but not by depletion of BRAG2 during the first 2 weeks after birth. In contrast, depletion of BRAG2 during postnatal weeks 4 and 5 reduced the number of AMPAR miniature events and compromised the quantal sizes of both AMPAR and NMDAR currents evoked at Schaffer collateral synapses. We conclude that both Arf6 activation through GluN2B-BRAG1 during early development and the transition from BRAG1- to BRAG2-dependent Arf6 signaling induced by the GluN2 subunit switch are critical for the development of mature glutamatergic synapses.

The maturation of glutamatergic synapses in the CNS is regulated by NMDA receptors (NMDARs) that gradually change from a GluN2B-to a GluN2A-dominated subunit composition during postnatal development. Here we show that NMDARs control the activity of the small GTPase ADP-ribosylation factor 6 (Arf6) by consecutively recruiting two related brefeldin A-resistant Arf guanine nucleotide exchange factors, BRAG1 and BRAG2, in a GluN2 subunit-dependent manner. In young cortical cultures, GluN2B and BRAG1 tonically activated Arf6. In mature cultures, Arf6 was activated through GluN2A and BRAG2 upon NMDA treatment, whereas the tonic Arf6 activation was not detectable any longer. This shift in Arf6 regulation and the associated drop in Arf6 activity were reversed by a knockdown of BRAG2. Given their sequential recruitment during development, we examined whether BRAG1 and BRAG2 influence synaptic currents in hippocampal CA1 pyramidal neurons using patch clamp recordings in acute slices from mice at different ages. The number of AMPA receptor (AMPAR) miniature events was reduced by depletion of BRAG1 but not by depletion of BRAG2 during the first 2 weeks after birth. In contrast, depletion of BRAG2 during postnatal weeks 4 and 5 reduced the number of AMPAR miniature events and compromised the quantal sizes of both AMPAR and NMDAR currents evoked at Schaffer collateral synapses. We conclude that both Arf6 activation through GluN2B-BRAG1 during early development and the transition from BRAG1-to BRAG2-dependent Arf6 signaling induced by the GluN2 subunit switch are critical for the development of mature glutamatergic synapses.
Although fast excitatory neurotransmission in the central nervous system is mediated by synapses containing both AMPARs 4 and NMDARs, developing synapses are often AMPAR-silent and rely on NMDARs only (1). NMDARs govern the functional maturation of glutamatergic synapses as well as activity-dependent alterations of their strength by changing membrane trafficking that affects the AMPAR content of the postsynaptic membrane (2).
NMDARs assemble as tetramers from two GluN1 and two GluN2/GluN3 subunits (2). A variety of four GluN2 subunits forms the basis for the functional diversity of the receptor. Whereas embryonic NMDARs contain GluN2B or GluN2D, GluN2A is increasingly expressed only during postnatal development (3)(4)(5) and incorporated into NMDARs of synapses receiving a strong activity (6). This developmental GluN2 subunit switch prevents premature formation of synapses but allows synapse maturation at sites of strong input (7)(8)(9). GluN2 subunits possess large intracellular domains and interact directly or indirectly with core components of the postsynaptic density (PSD) including PSD-95 family proteins and a number of signaling molecules including kinases, phosphatases, GTPases, and their regulators (10). The NMDAR channel mediates influx of calcium ions that act as second messenger molecules within the dendritic spine and are intimately coupled to neuronal plasticity processes (2).
Among the proteins strongly enriched in the PSD are BRAG1 and BRAG2, two closely related guanine nucleotide exchange factors (GEFs) for the small GTPase Arf6 (11)(12)(13)(14). BRAG proteins contain a Sec7 domain catalyzing the GDP-GTP exchange in tandem with a pleckstrin homology domain and an IQ-like motif that binds calcium-free calmodulin (15). Arf6 regulates membrane trafficking and actin cytoskeleton remodeling at the * This work was supported by the Deutsche Forschungsgemeinschaft (KO plasma membrane (16,17). Experiments in primary neuronal cultures indicated that Arf6 influences the maturation and maintenance of dendritic spines (18,19) and that a significant fraction of AMPARs recycles through Arf6-positive endosomes (20).
BRAG proteins have been implicated in neuronal development and plasticity (21). Mutations in the gene for BRAG1 were identified as the cause of X chromosome-linked intellectual disability in several families (22). Viral expression of BRAG1 in hippocampal slice cultures reduced evoked AMPAR currents depending on endogenous NMDAR activity (15). We found BRAG2 to be involved in different forms of long term depression (LTD) in the mouse hippocampus (13). LTD triggered through metabotropic glutamate receptors relied on a direct binding of BRAG2 to the AMPAR subunit GluA2 (13).
Here, we studied a link among NMDAR activity, BRAG-mediated Arf6 activation, and synaptic maturation. Our results show that BRAG1 and BRAG2 are consecutively recruited by NMDARs during development and that a loss of BRAG1 or BRAG2 affects the number and strength of excitatory synapses.
Adeno-associated virus (AAV) constructs were generated by inserting NheI-BstBI fragments encompassing the U6-promoter-shRNA cassettes for BRAG1-and BRAG2-RNAi into pAAV-CAG-GFP (Addgene) equipped with an appropriate linker immediately upstream of the CAG-GFP cassette. Viral particles of AAV serotype 9 were produced by the Viral Core Facility of the Charité Berlin using an iodixanol gradient for purification (27) and quantitative PCR for quantification.
Cell Culture-Media and supplements were obtained from Life Technologies unless otherwise stated. Primary cortical neurons were prepared from rat embryos at embryonic day 18 (E18). Primary hippocampal neurons were prepared from E18 rat and E16.5 mouse embryos. Cortices and hippocampi were isolated in Hanks' balanced salt solution, dissociated with 1 mg/ml trypsin (Worthington), and triturated with fire-polished Pasteur pipettes. The cell suspension was washed twice in growth medium (Neurobasal medium, B-27 serum-free supplement, penicillin (50 units/ml), streptomycin (50 g/ml), 2 mM GlutaMAX) and plated on poly-DL-ornithine hydrobromide (Sigma-Aldrich)-coated plates or coverslips. Cortical neurons were plated at a density of 700,000 cells/well in 6-well plates and 350,000/per well in 12-well plates, whereas hippocampal neurons were plated on 12-mm coverslips placed in 24-well plates at 75,000 cells/well. Half of the medium was exchanged on day in vitro 2 (DIV2), DIV4, DIV9, and then once a week.
BRAG-expressing Cell Lines-An HEK293 cell line expressing a 1154-amino acid isoform of BRAG1 (HEK-BRAG1) was generated following the same protocol as for HEK-BRAG2 (13).
GST Pulldown Assay-GST fusion proteins were expressed in Escherichia coli BL21(DE3) (Stratagene) and purified on glutathione-Sepharose beads (GE Healthcare). For in vitro interaction studies, membrane-free extracts of transfected HEK293 cells (PBS, 1% Triton TM X-100, 1ϫ Complete TM protease inhibitor mixture, pH 7.4) were precleared by incubation with a mixture of unloaded and GST-coupled glutathione beads and then incubated with beads coupled to the recombinant proteins for 2 h at 4°C. Beads were washed in PBS containing 0.1% Triton X-100.
HEK293 cells were starved for 1 h in ECS (140 mM NaCl, 1.3 mM CaCl 2 , 5.4 mM KCl, 25 mM HEPES, 33 mM glucose, pH 7.4) and then stimulated with 1 mM L-glutamate (Applichem) for 5 min in ECS. To test the importance of calcium influx, HEK293 cells were starved in ECS containing 1.3 mM CaCl 2 for 1 h, washed with ECS without CaCl 2 , and then stimulated with 1 mM L-glutamate in ECS with or without 1.3 mM CaCl 2 .
Arf6 activity was quantified by comparing background-corrected signal intensity densities of Arf6 pulldown (pd) and Arf6 total (t) bands on immunoblots. Band intensities were measured on the blot imager Fusion FX7 (Vilber Lourmat). Intensity values were obtained by Bio1D (Vilber Lourmat) and calculated with Microsoft Excel as pd/t ratios or effect percentages. GraphPad Prism version 5 for Windows was used for graphs and statistical analysis.
Results were expressed as means of active Arf6 ratios (pd/t) ϮS.E. from at least three independent experiments and six independent values. In the case that test groups were compared in pairs of treated and untreated samples, activity ratios are shown as effect percentages compared with the untreated control. Statistical significance was evaluated by two-tailed unpaired or paired t tests as indicated. Differences were considered significant (*) at p values of less than 0.05.
Fluorescence Internalization Assay-Hippocampal rat neurons were infected at DIV15 and used at DIV24/25. Neurons were treated under live conditions at 37°C, 5% CO 2 in their medium first for 10 min with mouse anti-GluA2 antibody (1:100). After three brief washings with conditioned medium, cells were stimulated with 100 M NMDA for 3 min, washed once with medium, and incubated for an additional 7 min. Cells were then mildly fixed with 4% paraformaldehyde for 5 min, washed with PBS, blocked with 1% BSA for 30 min at room temperature, and incubated with Cy5-conjugated anti-mouse antibody (Jackson ImmunoResearch Laboratories; 1:500) in 1% BSA to visualize surface GluA2. After washing with PBS, cells were fixed and permeabilized with a 9:1 mixture of methanol and MES solution (100 mM MES (AppliChem), pH 6.9, 1 mM EDTA, 1 mM MgCl 2 ) for 90 s at Ϫ20°C. Cells were again blocked for 30 min with 1% BSA, and chicken anti-GFP antibody (1:1,000) was applied for 1 h at room temperature. After washing with PBS, Cy3-conjugated anti-mouse antibody (Jackson ImmunoResearch Laboratories; 1:500) and Alexa Fluor 488-conjugated anti-chicken antibody (1:1,000) were applied for 1 h in 1% BSA. Coverslips were washed and mounted in Mowiol.
GFP-positive neurons were inspected on an SP5 confocal system (Leica Microsystems). The ratio of the mean intensities for Cy5 representing surface-remaining GluA2 and Cy3 representing intracellular GluA2 in segments of proximal dendrites was measured using ImageJ.
Electrophysiology in Berlin (Fig. 9)-Adeno-associated virus constructs expressing shRNA were injected unilaterally in the hippocampus of postnatal day 0 (P0) (BRAG1 RNAi or BRAG2 RNAi) and P21 (BRAG2 RNAi) mice. For viral injections in P0 mice, pups were briefly anesthetized with isoflurane. Viral constructs were directly injected in the hippocampal region at three different spots along the ventrodorsal axis 20 m apart. The virus was pressure-injected using borosilicate glass capillaries (World Precision Instruments) pulled with a Zeitz DMZ Puller (Martinsried, Germany). For stereotaxic injections in P21 mice, animals were anesthetized with ketamine/xylazine (100 and 10 mg/kg, respectively). Hippocampal injections were done using the following coordinates (relative to bregma): medial/lateral, 3.60 mm; anterior/posterior, 3.28 mm; dorsal/ ventral, 2.96 mm. Virus was injected at three spots 50 m apart, 0.25 l each using capillary micropipets (Drummond). At least 2 weeks after infection, mice were briefly anesthetized with isoflurane and decapitated. Brains were rapidly removed and transferred to cooled oxygenated artificial cerebrospinal fluid containing 87 mM NaCl, 26 mM NaHCO 3 , 50 mM sucrose, 10 mM glucose, 2.5 mM KCl, 1.25 mM NaH 2 PO 4 , 3 mM MgCl 2 , and 0.5 mM CaCl 2 . Horizontal hippocampal slices (300 m) were cut on a VT 1200 Vibratome (Leica Microsystems). Slices were then incubated at 34 -35°C in an interface-type storing chamber, and recordings were started after 1 h. All recordings were performed in artificial cerebrospinal fluid containing 119 mM NaCl, 26 mM NaHCO 3 , 10 mM glucose, 2.5 mM KCl, 1 mM NaH 2 PO 4 , 1.3 mM MgCl 2 , and 2.5 mM CaCl 2 at near-physiological temperature (ϳ34°C). All artificial cerebrospinal fluid solutions were equilibrated with 95% O 2 and 5% CO 2 ; osmolarity of artificial cerebrospinal fluid for recordings was maintained in the range of 290 and 305 mosM. Whole-cell patch clamp recordings were performed in a submerged recording chamber. For recordings, borosilicate glass electrodes (2-5 megaohms) were used filled with 130 mM KMeSO 3 , 10 mM HEPES, 20 mM KCl, 4 mM NaCl, 4 mM Mg-ATP, 0.5 mM Na-GTP, and 5 mM sodium phosphocreatine; pH of the intracellular solution was adjusted to 7.2 with KOH. To isolate AMPAR miniature excitatory postsynaptic currents (mEPSCs), 1 M SR 95531 (gabazine), 1 M tetrodotoxin, 50 M D-AP5, and 100 M cyclothiazide were added to the extracellular recording solution. Principal cells located in the pyramidal cell layer of the hippocampal CA1 area were identified using infrared differential interference contrast video microscopy and adequate fluorescence filter settings visualizing GFP-tagged neurons. Series resistance (R s ) was monitored throughout experiments; cells were rejected if R s was Ͼ20 megaohms. No R s compensation was used. Collected signals were digitized with 16-bit resolution (National Instruments, Austin, TX) and sampled at 3 kHz using Igor Pro (Wavemetrics, Lake Oswego, OR). For mEPSC analysis, data were filtered to 1 kHz. For mEPSC collection, signals were first detected automatically using the Igor plug-in Neuromatics and in a second step manually sorted by visual inspection to exclude false positive events.
EPSCs were evoked with biphasic electrical stimulation in stratum radiatum (ϳ100 -150 m from stratum pyramidale and shifted toward subiculum). In whole-cell voltage clamp with C fast and R s compensated for, EPSCs were recorded at room temperature in a submerged chamber from green fluorescent CA1 pyramidal cells (EPC10, HEKA, Patchmaster software) in saline (125 mM NaCl, 25 mM NaHCO 3 , 2.5 mM KCl, 1.25 mM NaH 2 PO 4 , 1 mM MgSO 4 , 2 mM CaCl 2 , 12.5 mM glucose, and 0.005 mM gabazine bubbled with 95% O 2 and 5% CO 2 ; ϳ300 mosM) with internal pipette solution (0.2 mM EGTA, 10 mM HEPES, ϳ125 mM cesium gluconate, 20 mM CsCl, 10 mM NaCl, 10 mM Na 2 -phosphocreatine, 4 mM Mg-ATP, 0.3 mM Na 3 -GTP, and 5 mM QX-314 chloride, pH 7.3; ϳ280 mosM). Control EPSCs were recorded from uninfected cells from the right hemisphere. AMPAR currents were recorded at Ϫ70 mV, and NMDAR currents were recorded at ϩ40 mV after wash-in of 5 M 2,3-dioxo-6-nitro-1,2,3,4-tetrahydrobenzo[f]quinoxaline-7-sulfonamide. Rise of NMDAR currents was determined by the slope of a 20 -80% linear fit normalized to the respective amplitude, decay of NMDAR current by a monoexponential fit, and paired pulse ratios with an interstimulus interval of 50 ms. The coefficient of variation (CV) was the ratio of standard deviation and average EPSC amplitude per cell (15-30 repetitions per cell). Mean EPSC was plotted against inverse squared CV (CV Ϫ2 ) assuming that (i) CV Ϫ2 ϭ n ϫ P r /(1 Ϫ P r ) with a constant average release probability of P r ϭ 0.3 (30) and (ii) mean EPSC is proportional to the number n of stimulated synapses and to the number of actually releasing synapses, which we call n r (n r ϭ n ϫ P r ). Therefore, quantal size q is proportional to the slope of a linear fit through the origin, i.e. q ϭ slope/(1 Ϫ P r ) because CV Ϫ2 ϭ n r /(1 Ϫ P r ) and mean EPSC ϭ q ϫ n r .
Statistical significances (*, p Ͻ 0.05) were assessed in GraphPad Prism 5.02 or Igor Pro 6.36 (Wavemetrics) via t test if not indicated otherwise.

Results
NMDARs Stimulate BRAG1 and BRAG2 in a Subtype-selective Manner in HEK293 Cells-To assess potential signaling from NMDARs to BRAGs, we co-expressed diheteromeric NMDARs consisting of GluN1 and either GluN2A or GluN2B and BRAG1 or BRAG2 in HEK293 cells and measured changes in Arf6 activity following stimulation with glutamate (Fig. 1A). GluN2B-containing NMDARs triggered an increase in the Arf6-GTP level through BRAG1. Vice versa, activation of GluN2A-containing NMDARs increased the Arf6 activity upon co-expression with BRAG2. In contrast, in cells expressing GluN2B and BRAG2 or GluN2A and BRAG1, the Arf6 activity was decreased by the glutamate treatment. Omission of calcium from the extracellular solution during stimulation prevented glutamate-triggered Arf6 activation, indicating that NMDAR-BRAG signaling requires calcium influx (Fig. 1B). BRAG1, BRAG2, and Arf6 were enriched at sites close to the plasma membrane in line with their regulation by surfaceexpressed NMDARs (Fig. 1C). Thus, it appears that the closely related Arf6 GEFs BRAG1 and BRAG2 are able to mediate calcium-dependent, subtype-specific functions of the NMDAR.
NMDAR-BRAG Signaling Depends on Binding Sites in the GluN2 C-terminal Tails-We used GST pulldown assays to test for interactions between the intracellular domains of GluN2 subunits and BRAGs (Fig. 2). A 200-amino acid region centrally located within the intracellular domain of GluN2A bound BRAG2 ( Fig. 2A). In a calcium-free buffer, this fragment of GluN2A as well as the corresponding region of GluN2B pulled down both BRAGs without preference. In contrast, in the presence of 100 M Ca 2ϩ , GluN2A pulled down less BRAG1 than Arf6-HA t pd t pd t pd t pd GluN2A/BRAG2 FIGURE 1. Ligand binding to NMDARs stimulates BRAG-mediated Arf6 activation in a subtype-selective and Ca 2؉ -dependent manner. A, subtype-selective stimulation of BRAG1 and BRAG2 by ligand binding to NMDARs in HEK293 cells. Shown are representative immunoblots of Arf6 GTPspecific pulldown assays from HEK-BRAG1 and HEK-BRAG2 cells expressing Arf6-HA, GluN1, and either GluN2A or GluN2B. Bars illustrate Arf6 activation calculated as the Arf6 GTP /Arf6 total ratio (pd/t) of cells treated for 5 min with 1 mM L-glutamate (glu) normalized to untreated controls (paired t test: HEK-BRAG1: GluN2A, *, p ϭ 0.0099, n ϭ 9; GluN2B, *, p ϭ 0.0033, n ϭ 7; HEK-BRAG2: GluN2A, *, p ϭ 0.043, n ϭ 6; GluN2B, *, p ϭ 0.025, n ϭ 6). B, NMDAR-BRAG signaling requires Ca 2ϩ . Shown are representative immunoblots of Arf6 GTPspecific pulldown assays from HEK-BRAG1 cells expressing Arf6-HA, GluN1, and GluN2B and from HEK-BRAG2 cells expressing Arf6-HA, GluN1, and GluN2A. Cells were stimulated by application of an extracellular solution containing L-glutamate (glu) with or without 1.3 mM Ca 2ϩ . Bars illustrate Arf6 activation calculated as the Arf6 GTP /Arf6 total ratio (pd/t) of cells treated with glutamate in the presence or absence of extracellular Ca 2ϩ normalized to untreated controls (paired t test: HEK-BRAG1: without (w/o) Ca 2ϩ , p ϭ 0.63, n ϭ 11; with Ca 2ϩ , *, p ϭ 0.0009, n ϭ 6; HEK-BRAG2: without Ca 2ϩ , p ϭ 0.88, n ϭ 8; with Ca 2ϩ , *, p ϭ 0.010, n ϭ 6). Error bars indicate S.E. C, subcellular distribution and expression levels of NMDARs, BRAGs, and Arf6 in transfected HEK-BRAG1 and -BRAG2 cells. NMDARs with HA-tagged GluN2 subunits and untagged Arf6 were expressed to compare the localization of GluN2B and BRAG1 or GluN2A and BRAG2 (left column). NMDARs with untagged GluN2 subunits and Arf6-HA (right column, as in A and B) were expressed to compare the localization of BRAG1 or BRAG2 and Arf6. The distribution of the stained proteins was similar upon expression of GluN2A in HEK-BRAG1 and GluN2B in HEK-BRAG2 cells (not shown).Western blots show the expression levels of GluN2 subunits, BRAGs, and Arf6-HA in homogenates of HEK-BRAG1 and HEK-BRAG2 cells transfected as in A and B. Scale bar, 10 m.  within GluN2 intracellular C-terminal tails. The scheme indicates the location of the protein fragments used for the interaction analyses within the C-terminal regions of GluN2A and GluN2B. NMDAR fragments shown to interact with BRAG1 or BRAG2 are depicted in green. Sequences of the shortest interacting fragments in single letter code of amino acids are shown below with conserved motifs in red and blue. Importance of the KTK/RTK motif (bold) was tested in Fig. 3.
GluN2B, whereas BRAG2 was pulled down exclusively by GluN2A (Fig. 2B). This finding suggests that calcium has an impact on the interaction between NMDARs and BRAG proteins. Further mapping of the BRAG binding sites within GluN2A and GluN2B revealed 40-amino acid fragments able to interact with BRAGs in the GST pulldown assay (Fig. 2, C and  D). Glutamate stimulation of NMDARs lacking these regions either decreased the Arf6 activity ( Fig. 3A) or had no effect on the Arf6-GTP level (Fig. 3B) in contrast to full-length receptors.
These data indicate that both a rise in the calcium concentration and interaction sites in the cytoplasmic region of the GluN2 subunits are required for the activation of BRAG1 and BRAG2. A comparison of the GluN2 sequences interacting with BRAG1 and BRAG2 revealed two short stretches of conserved amino acids (Fig. 2D). Interestingly, they are present in the opposite order in GluN2A and GluN2B. Mutation of KTKDN to AAADN in GluN2A and RTKEN to AAAEN in GluN2B prevented the glutamate-triggered Arf6 activation through NMDARs and BRAGs in HEK293 cells (Fig. 3, A and B). This result suggests that the amino acids 1081 KTK 1083 / 1138 RTK 1140 in GluN2A/2B are required for signal transduction from NMDARs to BRAGs.
BRAGs Mediate NMDAR-driven Arf6 Activation in Neuronal Cultures-It is well established that NMDARs are restricted to GluN2B or GluN2D in terms of GluN2 subunit expression at embryonic stages, whereas during postnatal development the contribution of GluN2A is continuously increasing (3)(4)(5). We therefore chose to analyze NMDAR effects on endogenous Arf6 in primary cortical neuron cultures at a young (DIV6 -8) and at a mature (DIV20 -22) stage. The expression levels of BRAG1, BRAG2, and GluN2A were much higher in mature than in young neurons, whereas GluN2B and Arf6 showed only minor changes (Fig. 9D). In young cultures, we found the Arf6 activity to be significantly higher than in mature cultures (Fig. 4A). The Arf6 activity of young neurons was decreased by both the NMDAR antagonist D-AP5 (Fig. 4A) and the NMDAR channel blocker MK-801 (Fig. 4B), suggesting an endogenous stimulatory NMDAR activity on Arf6 at this stage. We tried to isolate subtype-selective effects using 3 M ifenprodil to prevent activation of heterodimeric GluN1/GluN2B NMDARs or 300 nM Zn 2ϩ , which blocks preferentially GluN2A-containing NMDARs at this concentration (31)(32)(33)(34). The endogenous NMDAR signaling on Arf6 in young neurons was blocked by 3 M ifenprodil but not by 300 nM Zn 2ϩ , suggesting it was mediated by GluN2B (Fig. 4B). Next, we infected neurons at DIV2 with lentiviral vectors expressing short hairpin RNAs to deplete BRAG1 or BRAG2. Ifenprodil reduced the Arf6 activity in control (scrambled shRNA)-and BRAG2-RNAi-infected but not in BRAG1-RNAi-infected neurons at DIV7 (Fig. 4, C and D). In the absence of BRAG1, ifenprodil actually increased the Arf6 activity, suggesting a negative tone of GluN2B on Arf6 under these conditions. These data indicate that BRAG1 is mediating endogenous GluN2B-dependent signaling on Arf6 in young neuronal cultures.
In mature neurons, application of D-AP5 did not decrease the Arf6-GTP level (Fig. 4A), but stimulation with NMDA for 5 min caused an activation of Arf6 that was blocked by D-AP5 or MK-801 (Fig. 5A). The NMDA-triggered Arf6 activation in mature neuronal cultures was not affected by 3 M ifenprodil, indicating that it was not mediated by GluN1/ GluN2B receptors and that it might be caused by NMDARs containing GluN2A. Indeed, the NMDA stimulus did not lead to an activation of Arf6 in the presence of 300 nM Zn 2ϩ . Next, we infected neurons at DIV15 to deplete BRAG1 or BRAG2 in mature neurons and found that NMDA-triggered activation of Arf6 at DIV21 specifically relied on BRAG2 (Fig. 5, B and C). In the absence of BRAG2, NMDA caused a drop in Arf6 activity. Thus, in line with our experiments in HEK293 cells, NMDARs containing GluN2A signal through BRAG2. In summary, during the development of cortical neurons in culture, the two major NMDAR subtypes of the forebrain lacking or containing GluN2A triggered GDP/ GTP exchange on Arf6 through BRAG1 or BRAG2, respectively. Surprisingly, mature neurons infected for BRAG2 RNAi at DIV15 displayed an increased basal activity of Arf6 that was sensitive to ifenprodil in contrast to control-infected neurons (Fig. 6A). Thus, the knockdown of BRAG2 in mature neurons resulted in GluN2B-mediated Arf6 activation reminiscent of the situation in young neurons (Fig. 4B). To check whether the ifenprodil-sensitive Arf6 activity depended on BRAG1, we infected neurons with a vector for RNAi against both BRAG1 and BRAG2 at DIV15. Although this approach was less efficient in reducing the expression of BRAG1 and BRAG2 as compared with the respective single knockdowns (Fig. 4C), it entirely blocked NMDA-triggered Arf6 activation in mature neurons (Fig. 5B). Neurons carrying the knockdown construct for both BRAGs had a low Arf6 activity that was not sensitive to ifenprodil, suggesting that BRAG1 mediates GluN2B-triggered Arf6 activation in mature neurons lacking BRAG2 (Fig. 6A). Treatment of mature cultures with 300 nM Zn 2ϩ had a similar effect as the knockdown of BRAG2: an increased basal ifenprodil-sensitive Arf6 activity that was prevented by a knockdown of BRAG1 applied at DIV15 (Fig.  6B). These data imply that an interference with the GluN2A-BRAG2 pathway in mature synapses reinstalls GluN2B-BRAG1 signaling.
Together, our Arf6 activity measurements suggest that BRAG1 mediates tonic GluN2B-triggered Arf6 activation in young neurons in culture, whereas in mature neurons BRAG2 mediates GluN2A-triggered Arf6 activation upon NMDAR stimulation. This developmental shift appears to be reversible as depletion of BRAG2 or GluN2A blockade in mature neurons resulted in a high Arf6 activity mediated by GluN2B and BRAG1.

Effects of BRAG1 and BRAG2 Depletion on the Number and Size of Glutamatergic Synapses in Primary Neuronal Cultures-
We examined synapses in mature neuronal cultures by costaining the presynaptic marker VGLUT 1 and the postsynaptic marker PSD-95. The knockdown of either BRAG1 or BRAG2 reduced the number of overlapping puncta by about 20%. However, the knockdown of BRAG2 diminished the average total size of the overlapping puncta, whereas the knockdown of BRAG1 did not affect the size of synaptic contacts (Fig. 7). The reduced synapse size may be related to the increased Arf6 activity that occurred upon knockdown of BRAG2 but not upon knockdown of BRAG1 in mature neuronal cultures (Fig. 6, A  and C).
Effects of BRAG1and BRAG2 Depletion on NMDA-triggered AMPA Receptor Internalization in Primary Neuronal Cultures-NMDARs control AMPAR trafficking to fine-tune synaptic transmission. We asked whether BRAG1 or BRAG2 was involved in NMDA-triggered internalization of the principal AMPAR subunit GluA2 in dendritic segments of mature hippocampal neurons in culture (Fig. 8). NMDAR stimulation increased GluA2 internalization in control-infected neurons. RNAi-mediated knockdown of BRAG2, but not of BRAG1, interfered with this effect. However, in neurons lacking BRAG2, GluA2 internalization prior to stimulation was higher than in control-infected neurons, potentially occluding the NMDA effect. As the knockdown of BRAG2 in mature cortical cultures resulted in an increased Arf6 activity (Fig. 6A) and precluded NMDA-triggered Arf6 activation (Fig. 5C), these data are consistent with a role of BRAG2-mediated Arf6 activation in NMDA-triggered GluA2 internalization in mature neurons.
To test whether BRAG2 signaling exerts its impact in adult animals, we injected the viral construct for BRAG2 RNAi at P21 and recorded mEPSCs starting at P35. Now BRAG2 knockdown significantly reduced the frequency of synaptic events to levels seen in P16 neurons but did not alter their average amplitude ( Fig. 9C; ctrl, n ϭ 7 versus BRAG2-KD, n ϭ 8,; frequency, p ϭ 0.003; amplitude, p ϭ 0.92).
The hippocampal expression levels of BRAG1, BRAG2, and GluN2A but not of GluN2B and Arf6 increased between P14 and P36, although the differences were smaller than in rat cortical cultures between DIV8 and DIV22 (Fig. 9D). In addition, the GluN2A/GluN2B ratio was much higher in the hippocampus than in the cortical cultures.
We also compared the relative frequencies of miniature amplitudes after knockdown of BRAG1 or BRAG2 using an equal number of events per cell and condition to prevent overrepresentation of single cells. The relative frequency distribution of mEPSCs in young neurons infected for BRAG1 RNAi at P0 (Fig. 9A)  9C) revealed a shift to smaller amplitudes as compared with uninfected neurons (Kolmogorov-Smirnov test, p Ͻ 0.0001). We cannot exclude that the amplitude shift results in a reduction of the mEPSCs to a level below the detection limit in a subset of synapses, thereby contributing to the reduced mEPSC frequencies. In contrast, BRAG2 RNAi at P0 did not affect the relative frequency distribution of mEPSC amplitudes (Fig. 9B). Together, these results corroborate that signaling through BRAG1 and BRAG2 is critical at different points in development, although the expression of both BRAG1 and BRAG2 in mouse forebrain was detectable at P1 and increased steadily afterward (Fig. 9D). In addition, the altered amplitude spectra and the reduced mEPSC frequencies suggest that BRAG depletion reduced the strength and the number of synapses.
Effects of BRAG2 Depletion on the Properties of Evoked EPSCs in Mature Hippocampal CA1 Pyramidal Neurons-We next tested whether the increased GluN2B-mediated effects on Arf6 that we observed upon down-regulation of BRAG2 in cultured neurons (Fig. 6A) would manifest in a change of the synaptic NMDAR subunit composition in mature hippocampal CA1 pyramidal neurons of mice. To this end, we infected mice carrying loxP elements in the gene for BRAG2 (Iqsec1 fl/fl mice) (13) with a lentiviral Cre expression vector at 3-4 weeks of age. Two weeks later, isolated AMPAR and NMDAR currents were recorded from CA1 pyramidal neurons upon Schaffer collateral stimulation in acute slices (Fig. 10A). We did not observe any difference in the kinetic profile of NMDAR currents upon Cre infection ( Fig. 10B; ctrl, n ϭ 14 versus ⌬BRAG2, n ϭ 7; rise, p ϭ 0.38; decay, p ϭ 0.57). This result indicates that there was no increased participation of GluN1/N2B heterodimers in the synaptic NMDARs of cells lacking BRAG2 as this would have prolonged both the rise and decay of their currents (9,35).
The ratio between the peak amplitudes of AMPAR and NMDAR currents was increased to some extent in BRAG2 knock-out cells as compared with controls ( Fig. 10C; ctrl, n ϭ 14 versus ⌬BRAG2, n ϭ 7; p ϭ 0.048, Mann-Whitney test), suggesting a relative gain of AMPAR current and/or a relative loss of NMDAR current in mature CA1 neurons lacking BRAG2. A significant fraction of synapses on mature CA1 neurons contains functional NMDARs but lacks physiological AMPAR currents and is therefore called silent synapses (1, 30, 36 -38). The CV was significantly different between AMPAR and NMDAR currents in uninfected but not in BRAG2 knockout neurons ( Fig. 10D; AMPAR versus NMDAR; ctrl, n ϭ 14, p ϭ 0.0015; ⌬BRAG2, n ϭ 7, p ϭ 0.20), indicating a relative reduction in the number of silent synapses that may contribute to the increase in the AMPA/NMDA ratio. Paired pulse ratios were not altered by Cre infection (Fig. 10E; ctrl, n ϭ 6 versus ⌬BRAG2, n ϭ 9; p ϭ 0.62), suggesting no change in the release probability of glutamate. This allowed us to deduce the AMPAR and NMDAR quantal sizes from the relation of the mean EPSCs with their corresponding CV Ϫ2 , which served as a measure of the number of activated synapses (Fig. 10F). AMPAR and NMDAR quantal sizes were significantly reduced in cells lacking BRAG2 (ctrl, n ϭ 20 versus ⌬BRAG2, n ϭ 16; p ϭ 0.025 for AMPAR; ctrl, n ϭ 14 versus ⌬BRAG2, n ϭ 7; p ϭ 0.018 for NMDAR; linear regression F-test). Thus, CA1 neurons lacking BRAG2 had fewer AMPAR-silent synapses and lower AMPAR and NMDAR currents per Schaffer collateral synapse.

Discussion
Here we report interactions between NMDARs and BRAG proteins that mediate an activity-dependent regulation of neuronal Arf6-GTP levels. We previously found that the AMPAR subunit GluA2 stimulates BRAG2 in the course of LTD triggered by metabotropic glutamate receptors (13). Thus, BRAG stimulation appears to represent a general function of ionotropic glutamate receptors, ensuring a tight link between synaptic activity and BRAG-mediated Arf6 activation.
Ligand-bound NMDA receptors stimulate the GEF activity of BRAG1 and BRAG2 by calcium influx and by the large intracellular domains of GluN2B and GluN2A. The GluN2 subunits recruit a differential set of GEFs for the small GTPases Ras and Rac1 as well (39 -42). The fact that closely related GEFs associate with the C-terminal tails of GluN2A and GluN2B lends further support to the idea that gene duplication and paralog diversification increased the variety of postsynaptic signal transduction complexes (43). Quantitative mass spectrometry revealed that the numbers of BRAG1 and BRAG2 molecules per average PSD in adult rat cerebral cortex are in a similar range as that of NMDARs (around 10% of PSD-95), confirming their availability as postsynaptic binding partners (14). In the absence of BRAG1 or BRAG2, we frequently observed Arf6-GTP hydrolysis upon NMDAR stimulation. Thus, Arf6 is most likely regulated by synaptic pairs of GEFs and GTPase-activating proteins as shown for Rac1 (44). The NMDAR complex contains AGAP3, a GTPase-activating protein for Arf6 (45), that may be a candidate GTPase-activating protein partner of BRAG2.
We identified short regions within the C-terminal tails of GluN2A and GluN2B that mediate an interaction with BRAGs and are critical for signaling from NMDARs to BRAGs in vitro. These regions may serve as targets to manipulate NMDAR-dependent Arf6 activation in neurons. The interaction between GluN2 subunits and BRAGs may involve additional binding sites and/or proteins. Furthermore, PSD-95 family members may stabilize or coordinate NMDAR-BRAG complexes.
An increase in the intracellular calcium concentration was required for NMDAR-BRAG signaling in HEK293 cells, and calcium promoted selective interactions between NMDARs and BRAGs. The calcium sensor calmodulin interacts with the IQ-like motif of BRAG1 (15) and BRAG2 (data not shown) only in the absence of calcium. Calcium-triggered release of calmodulin induced a conformational change in BRAG1 (15). The calcium-sensitive calmodulin binding may modulate the interactions of BRAGs with GluN2 subunits and thereby contribute to the regulation of NMDAR-BRAG signaling.
Our experiments in primary cortical neurons revealed that GluN2B-BRAG1 signaling is replaced by GluN2A-BRAG2 signaling over time. In young but not in mature cultures, we detected tonic GluN2B-BRAG1 signaling that may contribute to the Arf6-dependent morphological maturation of dendritic spines (18,19) and/or to the relatively low plasticity threshold of young synapses (46). In mature neurons, NMDAR-depen- dent Arf6 activation relied on signaling through GluN2A and BRAG2 and was not induced by the endogenous excitation in the culture. Therefore, the incorporation of GluN2A into NMDARs affects not only the biophysical properties of their currents (4,9,35) but also restricts Arf6 activation to strong inputs. The tonic NMDAR-dependent Arf6 activation is shut down during development (Fig. 4A), although both GluN2B and BRAG1 are present at mature synapses (9,14,35). AMPAR-BRAG2 signaling is blocked by tyrosine phosphorylation (13), and a posttranslational modification may prevent GluN2B-BRAG1 signaling at the mature stage as well. However, BRAG2 RNAi in mature neurons induced a return to GluN2B-BRAG1 signaling. Thus, recruitment of BRAG2 may impair GluN2B-BRAG1 signaling, e.g. by competitive binding to triheteromeric  NMDARs. Application of 300 nM Zn 2ϩ reinstalled GluN2B-BRAG1 signaling in mature neurons as well, suggesting that GluN2A activation is required to maintain the low level of Arf6 activation. It will be interesting to see whether this phenomenon plays a role in vivo, e.g. during prolonged inactivity of synapses.
Measurements of AMPAR mEPSCs from hippocampal CA1 pyramidal neurons substantiated the importance of BRAG1 and BRAG2 at different stages of development. We found that BRAG1 plays a role in the maturation of synapses onto CA1 pyramidal neurons within the first 2 weeks of life. The mutations in the gene for BRAG1 identified in families with nonsyndromic X chromosome-linked intellectual disability result in a reduced GEF activity (22) and thus may affect GluN2B-BRAG1 signaling and the early postnatal development of synapses.
Although BRAG2 RNAi did not affect mEPSCs during the first 2 postnatal weeks, it resulted in a shift to smaller amplitudes and in a significantly reduced frequency of mEPSCs if applied at P21. In accord, knockdown of BRAG2 in mature neuronal cultures resulted in a decrease in the number as well as the size of glutamatergic synaptic contacts and elevated AMPAR internalization. Considering the tonic Arf6 activation through GluN2B-BRAG1 upon BRAG2 RNAi described above, these effects are in line with BRAG-dependent Arf6 activation mediating AMPAR internalization and synaptic depression (13,15). Furthermore, targeted deletion of BRAG2 in hippocampal CA1 pyramidal neurons of P21 mice reduced the fraction of silent synapses as well as the quantal sizes of evoked AMPAR and NMDAR currents. Together, the loss of BRAG2 compromised the synaptic connections onto CA1 pyramidal neurons.
The formation of functional synapses is thought to involve an initial GluN2B-dependent limitation of AMPAR trafficking to synapses and a subsequent relief from synaptic suppression upon incorporation of GluN2A (7)(8)(9). BRAG2 depletion reverted the drop in Arf6 activity induced by the developmental GluN2 subunit switch without altering the NMDAR subunit composition and caused synapse weakening. Therefore, we propose that the down-regulation of the Arf6 activity induced by the incorporation of GluN2A contributes to the formation of mature synapses.
The coordinated transition of GluN2B-BRAG1 to GluN2A-BRAG2 signaling may be critical for synaptic selection because it ensures that the tonic synapse-suppressing Arf6 activity is switched off only at those connections with a strong or correlated neuronal activity, whereas others get eliminated. Tonic Arf6 activation may also contribute to metaplasticity based on an increased participation of GluN2B in NMDARs (46 -49).
We previously identified a pathway through metabotropic glutamate receptors and AMPARs to stimulate BRAG2 and found that neither metabotropic glutamate receptor-nor NMDAR-dependent LTD could be induced in mature hippocampal CA1 neurons lacking BRAG2 (13). The GluN2A-BRAG2 signaling reported here may provide an explanation for the block of NMDAR-dependent LTD upon BRAG2 deletion. Conversely, the suppression of synaptic currents caused by the absence of BRAG2 may have prevented further depression upon LTD-related stimuli.
In summary, NMDARs, via their GluN2 subunits, consecutively stimulate BRAG1 and BRAG2 in the process of synapse maturation. It will be important to address the roles of these two pathways in homeostatic plasticity as well as in neurological disorders.
Author Contributions-M. N. E. designed, performed, and analyzed the experiments shown in Figs. 1, A and B, 2, and 3 Fig. 9, A-C. J. P. designed, performed, and analyzed the experiments shown in Fig. 10. D. S. and G. K. designed experiments and analyzed results. H.-C. K. conceived and coordinated the study, generated DNA constructs, and wrote the paper. All authors reviewed the results and approved the final version of the manuscript.